Monitoring Trends in Violence

Abstract

Reliable and valid indicators of assault are required to effectively monitor population trends and ensure that resources are targeted effectively. Trends in assault, reported by the media, based on crime statistics, or on victim surveys, are substantively affected by extraneous factors. In 2006, Estrada offered up solutions to the difficulties posed by crime statistics and victim surveys, by proposing the development of indicators based on hospital discharge data, albeit with identified limitations. This article is a response to Estrada’s proposition, and works through each of Estrada’s identified limitations of hospital discharge data. Potential problems with Estrada’s suggestions are highlighted in our article and solutions, based on the current evidence, are proposed.

Keywords

community violence criminology violence exposure

Introduction

The World Health Organization (WHO) defines violence as “the intentional use of physical force or power, threatened or actual, against oneself, another person, or against a group or community, that either results in or has a high likelihood of resulting in injury, death, psychological harm, maldevelopment, or deprivation.” (Violence Prevention Alliance, 2012). The term includes three subtypes according to the victim–perpetrator relationship: self-directed violence (self-abuse/suicide), interpersonal violence (family, interpartner violence, and community violence), and collective violence (social, political, and economic violence; Violence Prevention Alliance, 2012). Other definitions also exist. For example, in the United States Federal Bureau of Investigation’s Uniform Crime Reporting programme, “violent crime” is encapsulated in the offenses which involve force or threat of force (murder, forcible rape, aggravated assault, robbery; Federal Bureau of Investigation, 2010). The Oxford Dictionary provides two definitions for violence: behavior involving physical force intended to hurt, damage, or kill someone or something; or strength of emotion or of a destructive natural force (Oxford Dictionaries, 2012).

The discussion presented in this article is concerned with interpersonal violence as defined by WHO. The main advantage of the WHO definition, as opposed to other definitions of violence highlighted above, is that, because the majority of developed countries use International Classification of Diseases (ICD; a WHO coding frame) for coding nature and circumstances of injury hospital discharge and mortality records, discussions about the relative merit (or lack thereof) of this coding scheme are applicable on an international scale. Within WHO’s International Statistical Classification of Diseases and Other Health Related Problems, 10th revision (ICD-10), interpersonal violence is termed “assault” and is used to classify “injuries inflicted by another person with the intent to injure or kill, by any means” (National Centre for Classification in Health, 2000a). This article is concerned with the reliable monitoring of assaultive violence, herein simply termed assault.

To (a) adequately define the problem, (b) effectively monitor the application of population based interventions, and (c) ensure that resources are targeted appropriately, it is important that policy makers, injury prevention practitioners and researchers have access to assault statistics that are not misleading. In 2006, Estrada contributed to the international debate concerning the development of valid indicators for measuring trends in assault and associated injury over time by highlighting the potential usefulness of hospital discharge data for monitoring the incidence of assault in the community (Estrada, 2006).

As with other researchers who have written on this area, Estrada expressed concern that trends in assault reported by the media and based on crime statistics were substantively affected by extraneous factors. For example, the number of assaults reported to the police “are sensitive to shifting perceptions of what constitutes an act of violence and how such acts should be controlled” (p. 487). Estrada contended that there were variations across different groups and situations concerning what constitutes an assaultive act (Estrada, 2006). What may result in a warning on the rugby field may result in overnight incarceration if it occurred in a city outside a public bar. Changes also occur over time in what is considered to be an assaultive act and how these should be controlled (Estrada, 2001).

Estrada and others (e.g., Weatherburn, 2011) have also identified victim surveys as an alternative method of deriving population estimates of the level of assault in the community. Victim surveys typically include a representative sample of a defined population. If the questionnaire content is consistent and the sample representative of the population from which it was drawn, these surveys can provide an estimate of the prevalence of crime victimization over time (Weatherburn, 2011). Victim surveys represent the experience of those who participate in the survey. However, as with any survey, they suffer from selective attrition, under- and over-reporting, and the potential for the reporting to be influenced by changing social norms. They may also be more sensitive measures of some types of assaultive acts (work-related violence) than others (child abuse; Weatherburn, 2011). As such, victim surveys have limited use for producing trends in assault over time.

Shepherd and Sivarajasingam have recommended using hospital emergency department (ED) data as an objective measure of assault-related harm to inform local prevention efforts even though they acknowledged that this data also contains some threats to validity (Shepherd & Sivarajasingam, 2005). In a subsequent commentary, on Langley queried their recommended use of ED data over police data: “indicators based on police and health data both run a high risk of being misleading . . . could the authors advise readers why they advocate for emergency department in preference to police data?” (Langley, 2006, p. 208). Whether a person attends ED, should they have sustained an injury, will depend on the severity of the injury received during an assaultive event, the proximity to the ED, and other factors affecting ease of access (Shepherd & Sivarajasingam, 2005). In contrast, police statistics largely represent police activity rather than the actual incidence of assault in a community. With finite resources, the number of assaultive events that police will be able to attend (and therefore for which data will be collected) will be dictated by whether there exists other, more important events occurring at the same time.

In his review of different data sources for measuring trends in assault, Estrada recommended the use of hospital discharge data. The claimed advantage of this data source over others was that “hospital admissions constitute a measure that is not directly determined by conceptions of violence and of how it should be controlled, nor by the propensity to report an event to the police” (Estrada, 2006, p. 487). Estrada acknowledged several limitations in hospital discharge data, and proposed some solutions. While we agree with Estrada in respect of the limitations of data sources, we suggest that some of his (and others’) proposed solutions to these limitations introduce their own problems.

This article focuses on indicators of serious assault-related injury incidence, and contributes to the debate on what constitutes a valid indicator for measuring trends in assault over time. We work through each of Estrada’s identified limitations of hospital discharge data, and the solutions proposed by Estrada and other researchers in this area. We highlight what we consider problems inherent in their approach and present work that we have conducted that further illuminates the issues. Finally, we present what we consider the best solution, based on current evidence. Although the research presented in this article is drawn from New Zealand hospital discharge data, as highlighted above, because of the use of the ICD coding scheme in the majority of developed countries, we believe our proposed solutions have international relevance.

To place this discussion in context, the following sections introduce the New Zealand National Minimum Data Set (NMDS) of Hospital Discharges, from which the data presented in this article are drawn. In addition, we introduce our case definition for assault-related injury and describe the New Zealand national definition of a serious nonfatal injury.

Serious Nonfatal Assault Injuries in New Zealand

In 2003, the New Zealand Injury Prevention Strategy was launched as way of ensuring a coordinated approach across government agencies to reduce the impact of fatal and nonfatal injuries. Within the strategy, six priority areas were defined and included assault (Accident Compensation Corporation, 2003). To monitor the implementation of the strategy, indicators were developed for fatal and serious nonfatal injury (as defined below) for “all injury” and the six priority areas. Each year, trends (frequencies and age standardized rates) in fatal and serious nonfatal injury are published. Figure 1 shows the age standardized rates for the serious nonfatal assault indicators. The indicators are considered official measures of New Zealand’s serious injury incidence rate (Statistics New Zealand, 2010).

Figure 1.

Indicator for serious nonfatal assault: Age standardized rates for New Zealand 1994-2009.

The New Zealand NMDS of Hospital Discharges

The NMDS is a central repository for all inpatient and short-stay (day-patients and ED) public (since 1993) and private (since 1997) hospital discharges in New Zealand. It is maintained by the Ministry of Health (MoH). Hospital discharge data are transferred on to electronic patient management systems at individual hospitals by trained hospital coders. Diagnosis and external cause of injury on the discharge records are coded according to the ICD, 10th revision, Australian Modification (ICD-10-AM). This is a clinical modification of the standard WHO ICD-10 (National Centre for Classification in Health, 2002). Electronic records are then submitted to the MoH. The NMDS is used for policy formation, performance monitoring, research, review and funding purposes (Ministry of Health, 2012).

Our unit (the Injury Prevention Research Unit [IPRU]), within the University of Otago, maintains current and historical NMDS data for research and for producing national injury statistics. Annually, all hospital discharges with an ICD-10-AM external cause of injury code (e-code) are requested from MoH. According to ICD-10 coding rules, all cases with an injury related diagnosis (e.g., fractured femur, burn, poisoning) should attract an e-code (environmental events or circumstances that describe the cause of the injury, e.g., fall from one level to another due to collision with, or pushing by, another person). However, e-codes may also be associated with other diagnoses (e.g., sexually transmitted disease arising from assault) and so are not specifically related to injury cases. Therefore further refinement of the case selection is required to extract injury cases.

Case Definition of Assault-Related Injury

Internationally, the most commonly accepted operational definition of injury is those pathologies in the “Injury” chapter of the ICD diagnostic codes. For our national assault indicators, injury events in the NMDS, were identified using the ICD-10-AM (National Centre for Classification in Health, 2000a) diagnosis codes. Those cases with a principal diagnosis in the range S00-T78 were considered injuries¹ (approximately 60,000 to 80,000 injury discharges per year for the period 2001-2010). Assault-related injury, as recorded in the NMDS, was identified by ICD-10-AM e-codes. Those injury events with a first listed e-code in the range X85-Y09 were included (5% of injury discharges; n = 3,583 [2001] to n = 4,971 [2010]; National Centre for Classification in Health, 2000a). These codes include, for example, assault with a knife, by firearm discharge, and by bodily force.

Determining Seriousness of Assault Injury

For the New Zealand national indicators, developed by the IPRU, seriousness was defined in terms of threat to life. A serious nonfatal injury case was defined as one that is hospitalized and has an ICD-based Injury Severity Score (ICISS) of ≤0.941 (Stephenson, Langley, & Civil, 2002). This is equivalent to selecting patients whose injuries give the patient a survival probability of 94.1% or worse (a probability of death [at admission] of at least 5.9%). The reason for the level of precision of this cut off lies in the face validity of the threshold for determining threat to life. The injury diagnoses captured using different threshold levels were compared. A threshold of 5.9% captured most of the events of interest, and which were perceived to have a high probability of admission across all hospitals in New Zealand (Cryer, Langley, & Stephenson, 2004). The threshold for the previous revision of ICD (ICD-9-CM(A)) was also based on empirical work that showed a distinct split in trends for ICD-9-CM(A) ICISS ≤ 0.96 relative to ICISS > 0.96. (See Langley, Stephenson, & Cryer, 2003). An ICISS score less than or equal to 0.96 for the period that ICD-9-CM(A) was used to code New Zealand hospital discharges (1994-1999), was assessed as equivalent to an ICISS score of ≤ 0.941 for hospital discharges coded to ICD-10-AM (2000 onwards).

The ICISS method is based on previously calculated diagnostic-specific survival probability (DSP) for each individual ICD injury diagnosis code. These are calculated from a reference set of data as the ratio of the number of patients with a specified injury code who survive to the total number of patients assigned that diagnosis code. Each patient’s ICISS score (estimated survival probability) is the product of the probabilities of surviving each of their injuries individually, that is, a product of the relevant DSPs (Stephenson et al., 2002).

Among first admissions for injury in the NMDS, cases with an ICISS score ≤ 0.941 represent around 15% of all injury discharges from hospital. For assault-related injury discharges, those considered serious range from 19% in 2001 to 20% in 2010. This severity threshold includes the following injuries: fracture of the neck of femur, intracranial injury (excluding concussion only injury), injuries to the nerves and spinal cord at neck level, multiple fractures of the ribs, asphyxia, and injury diagnoses of similar or greater severity (Cryer & Langley, 2006).

The Problems With Hospital Discharge Data

In respect to the use of hospital discharge data, Estrada raises the following issues:

changes in treatment practice affecting:

Probability of admission;

Length of stay in hospital;

missing external cause of injury codes;

changing reporting practice;

changes in ICD versions affecting trends.

We will consider each of these issues separately.

Results

Changes in Treatment Practice

Probability of Admission

To counteract the effect of changing treatment/admission practices on measures of assaultive injury incidence, Estrada analyzed the Swedish hospital discharge register by causes and diagnoses considered sufficiently serious to make inpatient treatment highly likely (Estrada, 2006). Five types of injuries were considered: (a) injuries from guns, (b) injuries from knives, (c) concussions, (d) fractures, (e) all other injuries.

While well-intentioned, the method suffers from several limitations. For example, the weapon used in a violent act may not always be recorded; or changes in ICD coding may increase the likelihood of a weapon being recorded. In many localities concussion injuries are no longer routinely admitted to hospital, while the advent of fracture clinics reduce the likelihood of fracture related injury events being treated within an inpatient ward.

Instead of subjectively determining which diagnoses be included for hospital based assault surveillance, it is possible to identify cases based on a threat to life (TTL) threshold (Cryer et al., 2004), such as that described above.

As Estrada points out, the goal should be to identify cases with injury diagnoses that have a high probability of admission. It is these cases that will almost always be admitted to hospital and, as such, indicators based on them reduce the likelihood that extraneous factors, such as admission policies, will influence trends over time.

Work recently completed suggests that high TTL cases (ICISS score ≤ 0.941) have a high probability of admission (Cryer et al., 2011). In an international investigation with participation from Australia, Canada, Denmark, Greece, Spain, and the United States, the diagnoses captured by the ICISS threshold described above have been investigated for their probability of admission (Cryer et al., 2011). Collaborators in each of the participating countries had access to ED data from which (through linkage or direct recording) it was possible to determine which cases were subsequently admitted to an inpatient ward for at least 3 hr.

Probability of admission estimates for diagnoses with survival probabilities at the ICD-10 four-character level of ≤0.941 were investigated. Such diagnoses, that also had a high probability of admission, accounted for 97% of serious threat to life cases from the New Zealand data (Cryer et al., 2011). For assault-related hospital discharges, cases with a high probability of admission and DSP ≤ 0.941 accounted for 96% of serious threat to life assault cases in the New Zealand data.

The probability of admission project indicated that, with the exception of a small number of diagnoses, the current threat to life thresholds used for assault indicators go a long way to satisfying the goal of capturing only those diagnoses with a high probability of admission. We concluded that, the indicators were shown empirically to be valid to an acceptable level, and, on the whole captured those events with a high probability of admission.

Using the data collected from the probability of admission project (international diagnosis specific probabilities of admission), we attempted to investigate the relative probability of admission of the case groupings specified by Estrada. This was precluded for the first two groupings as they were defined by the cause of the injury, rather than the diagnosis of the injury received (wounds caused by guns and knives). Concussion incidents were recorded by two of the four countries for which hospital data were coded according to ICD-10. For these countries, the probability of admission for concussion injuries ranged from 0.34 (95% CI [0.33-0.35]) to 0.38 (95% CI [0.36-0.40]). Estrada provided a very general definition of the injury diagnoses included in the fourth category “acts of violence resulting in fractures (e.g., to the nose, jaw, ribs, arms)” (p. 490). For this group, we included fracture of the mandible; multiple fractures involving skull and facial bones; fractures of other skull and facial bones; fracture of skull and facial bones, part unspecified; multiple fractures of ribs; fracture of both ulna and radius; multiple fractures of the forearm. The average probability of admission for these combinations of diagnoses ranged from 0.47 to 0.67 (Cryer et al., 2011). It was not possible to estimate the probability of admission for the fifth category “Injuries not covered by these four categories have been grouped together in a fifth,” (p. 490). Where we were able to provide empirical results, these show that the Estrada categories, aimed at identifying groups with high probabilities of admission, are problematic.

Length of Stay in Hospital

Estrada equates length of stay in hospital to the relative seriousness of the injuries under investigation (p. 491). However, he then goes on to acknowledge that there has been variation in the trends for length of stay for assault injuries over time (pp. 496-498).

Our previous work has shown that, for all injury hospitalizations, there was a reduction in the median length of stay in New Zealand’s hospitals between 1989 and 1998 (interpolated medians 2.9 and 2.0, respectively; Cryer, Gulliver, Langley, & Davie, 2010). Of more interest within the context of this article, however, are the trends for length of stay for serious TTL nonfatal injuries. The calculation of our serious nonfatal threshold can only be applied to New Zealand hospital data for the period from 1994 onwards (Cryer et al., 2004). For all serious TTL nonfatal hospitalizations, from 1994 to 1998, there was a significant reduction in the median length of stay in New Zealand hospitals (interpolated median 1994 = 15.8; 1998 = 13.2). For assault-related serious nonfatal injuries, the interpolated median length of stay in 1994 was 6.4 reduced to 5.5 in 1998, although this was not statistically significant.

In New Zealand, for all injury hospitalizations in the period 1989 to 1998, the proportion with 7 or more days stay reduced from 30% to 21% (Cryer et al., 2010). For serious TTL nonfatal hospitalizations for the period 1994 to 1998, the proportion with 7 or more days stay reduced from 79% to 74%. For assault-related serious nonfatal hospital discharges (1994-1998), the proportion of cases with a length of stay of 7 or more days stay reduced from 50% to 40%. As Estrada indicates, the decrease in hospitalization times cannot solely be attributed to a reduction in the level of serious assault (Estrada, 2006). In fact, between 1994 and 1998, the frequencies and age standardized rates of serious nonfatal assault cases in New Zealand rose from 195 to 391 and 5 per 100,000 to 9 per 100,000, respectively (Statistics New Zealand, 2011b). Instead, during this time period, the proportion of cases with a length of stay of 7 days or more reduced, while the proportion with 3 to 7 days stay increased from 28% to 36%.

Estrada recommends removing those with 0 days stay, but only for a defined period of time (1974-1989). The basis for this decision is that it allows for compensation for the possibility that “events that had previously resulted in short hospitalisation times did not result in any kind of admission in the 1990s” (p. 500). Having a variable case definition for injury incidence based on administrative data sets, when the goal is to monitor trends, can be a dangerous practice. As we have previously shown elsewhere, the inclusion or exclusion of those with 0 days stay can have a profound effect on the trends in injury (Langley, Stephenson, Cryer, & Borman, 2002). In New Zealand between 1989 and 1993 there was a steady increase in the age standardized rate of 0 day stay patients recorded in the NMDS (from around 80 per 1,000 injury discharges in 1989 to around 180 per 1,000 injury discharges in 1998 [Langley et al., 2002]). It is hypothesized that this was driven by extraneous factors rather than changes in the incidence of injury. To reduce the impact of extraneous factors, it was recommended that hospitalized injury incidence should be consistently identified as those cases with a first listed diagnosis in the range S00-T78 (ICD-10), with an external cause of injury code in the range V01-Y34, and with a length of stay of 1 day or more (Langley et al., 2002). The subsequent focus on serious injury, defined using a recognized and validated severity scale and the application of a severity threshold as outlined above, for the New Zealand national injury indicators further removes the influence of extraneous factors.

Missing External Cause of Injury Codes

Estrada stated that there was a problem with missing data within Swedish hospital discharge records (1% for age, length of stay, gender, and diagnosis; 5% for external cause code; Estrada, 2006, p. 490). The proportion of injury hospital discharge events with missing data is likely to differ by country. In New Zealand in a 6-month period in which 152,977 injury events were recorded, there were only six discharges with missing e-code information (Angela Pidd Ministry of Health, Personal Communication, July 2012).

Davie and colleagues conducted an investigation into the accuracy of injury coding in New Zealand for the first 4 years of the implementation of ICD-10-AM. Of the randomly sampled records (n = 1,800), 81 were assault-related hospital discharge events. Of these 1,800, 25% (16-36) had the first, second or third character recorded incorrectly (Davie, Langley, Samaranayaka, & Wetherspoon, 2008). It is in these first three characters that errors in recording may result in an injury event being classified as assault when it is unintentional or vice versa.

Due to inaccuracies in coding the external cause of injury, it is possible that the number of assault cases may be under- or over-counted at any given time, and that the direction of bias may change. Inaccuracies in coding were more evident when there was unclear documentation by hospital clinicians (Davie et al., 2008). Hence, given that ICD coding rules state that assault external cause of injury codes should not be used unless there is clear documentation of such by clinicians (National Centre for Classification in Health, 2002), it is likely that any bias would result in an underreport of hospitalized assault.

Changing Reporting Practice

While previous validation has established the worth of the ICISS threshold for determining threat to life (Stephenson et al., 2002), the New Zealand national estimate of the incidence and rate of serious TTL nonfatal assault was described as “provisional” because of the potential for social norms to impact on the likelihood of an injured person reporting, or health staff determining, that the injury was the result of an assault (Cryer et al., 2004). Therefore we sought to determine whether, given a true assaultive injury presented at, and was admitted to, hospital with a high risk diagnosis, the likelihood of reporting an assault event had changed over time (Gulliver, Cryer, & Davie, 2009).

We identified sentinel diagnoses for assault that satisfied two criteria:

Those diagnoses that most frequently occurred among cases coded to assault.

From these diagnoses, those which have a high proportion of assault as the external code.

This resulted in the identification of just two diagnoses: S024 (Fracture of the malar and maxillary bones) and S0260 (fracture of the mandible, part unspecified).

Although in this New Zealand study we intended to produce sentinel diagnoses for males and females separately, insufficient numbers of serious nonfatal assault cases for female precluded this. Therefore, to provide some insight into the variability of the assault indicator for other hospital discharges for groups with relatively small numbers of assaults resulting in serious injury (such as females and young children), we also conducted the analysis using the aggregate of all of the diagnoses identified as occurring most frequently in the age-sex breakdowns among cases coded to assault and for which there was a high proportion with an assault e-code (herein referred to as “high frequency diagnoses”).

For each sentinel diagnosis (S0260 and S024), and for the group of high frequency diagnoses, we investigated the proportion coded to assault for serious nonfatal hospital admissions.

Table 1 presents the proportion of assault cases for the population of serious nonfatal hospitalizations with either of the sentinel diagnoses between 2001 and 2007. There was no significant difference between years in the proportions (χ² (6df) = 2.8, p = 0.8), nor was there any detectable linear trend over time (χ² (1df) = 0.76, p = 0.4).

Table 1.

New Zealand Serious Nonfatal Hospital Discharges With a Principal Diagnosis of “S024” or “S0206.”

	Year of Discharge^a,b
Assault External Cause Code	2001	2002	2003	2004	2005	2006	2007
No	50	43	42	57	76	82	63
%	41	36	35	41	40	43	41
Yes	71	77	77	82	114	110	90
%	59	63	65	59	60	57	59
Total	121	120	119	139	190	192	153

Difference over years in the proportions assigned an assault external cause code, χ² (6df) = 2.8, p = .8. ^bChanged reporting of assault external cause code, linear trend over time, χ² (1df) = 0.76, p = .4.

The proportion of assault cases for the population of serious TTL nonfatal hospitalizations with any of the high frequency diagnoses is presented in Table 2. Again, there was no difference between the years in the proportions with an assault external cause code (χ² (6df) = 11.71, p = 0.1), nor was there any significant linear trend across the years identified (χ² (1df) = 0.1, p = 0.7).

Table 2.

New Zealand Serious Nonfatal Assault Hospital Discharges With “High Frequency” Diagnosis^a.

	Year of Discharge
High Frequency Diagnosis	2001	2002	2003	2004	2005	2006	2007
No	461	483	451	462	539	605	616
%	68	70	70	64	67	66	68
Yes	214	203	195	260	269	310	306
%	32	30	30	36	33	34	32
Total	675	686	646	722	808	915	915

“High frequency” diagnoses were: “S0085” superficial injury of other parts of head; “S010” open wound of scalp; “S022” fracture of nasal bones; “S023” fracture of orbital floor; “S024” fracture of malar and maxillary bones; “S0260” fracture of the mandible, part unspecified; “S028” fracture of other skull and facial bones; “S065” traumatic subdural hemorrhage. ^bDifference over years in the proportions assigned an assault external cause code, χ² (6df) = 11.71, p = .1. ^cChanged reporting of assault external cause code, linear trend over time, χ² (1df) = 0.1, p = .7.

From the results presented, we found no systematic changes in the recording of assault, during the period 2001 to 2007, for cases that result in serious TTL nonfatal injury. This led us to conclude that, if there had been changing social norms in respect to identifying and reporting that an event was assaultive, it had no impact on New Zealand’s serious TTL nonfatal assault indicators.

Changes in ICD Revisions Affecting Trends

Estrada highlights changes to Swedish hospital discharge coding since 1987 when the Swedish register first covered the whole country. From 1987 to 1996, the Patientregistret used ICD-9 to code the nature and external cause of injury. In 1997, ICD-10 was implemented. Estrada suggests that “substantial shifts occurring with respect to individual diagnoses in connection with these changes in the register’s classification system . . . should be interpreted with caution” (Estrada, 2006, p. 489). We agree with the caution promoted by Estrada on this point. Figure 1, taken from New Zealand’s 2010 Chartbook of Serious Injury Outcome Indicators, shows that there was a substantial increase in estimated assault-related serious threat to life injury rates between 1998 and 2001 in New Zealand (Gulliver, Cryer, & Davie, 2010). New Zealand implemented ICD-10 over 2 calendar years: 1999 to 2000.

Aside from the dramatic difference in rates that coincide with the change from the 9th to 10th revision of the ICD coding scheme, there are also more subtle changes that will impact on case capture if cases are defined on the basis of the mechanism of injury (as proposed by Estrada). Figure 2 presents the trends in knife and gunshot wounds admitted to New Zealand hospitals between 2001 and 2010. Although this figure starts toward the end of the series presented by Estrada, there are similarities that provide insight into the Swedish data.

Figure 2.

Trends in selected mechanisms and diagnoses for all New Zealand assault hospital discharges*.

Between 2001 and 2004, New Zealand hospital coders were using the first and second editions of ICD-10-AM (National Centre for Classification in Health, 2000b). In the 2004/2005 financial year, the third edition was implemented (National Centre for Classification in Health, 2002). The release of the third edition, which included additional coding options for sharp weapons used in a violent event that were not available in the previous editions (X99.0-X99.9), resulted in a sudden and dramatic increase in the number of violent injuries attributed to knives (Figure 2). The same discontinuity in data could be seen in Figure 4 of Estrada’s article between 1997 and 1998, in his case, attributed to the ICD-9 to ICD-10 coding change. The rapid increase (which coincided with a rapid decrease in fractures) in violent injuries associated with knives (preceded and followed by periods of relatively stable incidence), are as Estrada suggests, indicative of the effects of changes to coding frames.

In summary, one should first look for explanations based on coding changes if there are sudden changes in the trend. In addition, case definitions that minimize the impact of coding changes should be used as the basis of indicators for monitoring trends in assault injury.

The Presentation of Trends

Throughout his article, Estrada presents trends in hospitalized assault as frequencies. While frequencies provide useful information concerning the population burden of assaultive injuries, they provide little information concerning an individual’s risk. With the changing demographics of most western populations (an increased average age), age standardized rates are necessary to compare risk over time. It may be that some of the reduction in assaultive injury described by Estrada was driven by a reduction in the proportion of the population aged 15 to 35 years and an increase in the proportion aged above 65 years. In other words, a migration of the age of the population from age groups of highest risk to those of relatively low risk.

Monitoring Trends in Violence: A Possible Solution?

Since 2006, the IPRU has been using the NMDS to produce indicators for serious nonfatal assaultive injuries. These have used retrospective hospital discharge data, typically spanning the years from 1994. While there have been a number of service delivery and admission policy changes over this time, not least the inclusion of ED attendances in the NMDS, the use of the ICISS threshold to identify eligible cases has minimized the impact of these policy changes on the trends in the indicators (Langley & Gulliver, 2012). This is not to suggest that we wish to minimize the influence of all policy changes on assault-related injury incidence. For example, there are policy changes aimed at level of policing, carrying of weapons, licensing of firearms, that should influence trends in hospitalized assault, and for which we want to capture their impact using our indicators. However, there are other policy changes that affect the likelihood of admission, the length of stay in hospital and the capture of cases on data systems. For these, we want to minimize their effect on indicator trends.

Although our approach is the best possible, given the limits of the New Zealand national data, and of scientific progress, there are, however, limitations to the approach we have used for the generation of the New Zealand assault indicators, described earlier in this article. While the ICISS method will reliably capture the majority of serious nonfatal hospitalized injuries resulting from a physical assault, those injuries that may be disabling but not high threat to life (such as those that result in psychological or emotional but not physical injuries) will not be captured using this method. Also, use of the case definition described only includes those who have a principal diagnosis in the “Injury” chapter of ICD-10. While the number of serious nonfatal assaultive injury hospitalizations that do not have a principal diagnosis in the injury chapter may not be substantial at the population level (Cryer, Gulliver, Langley, Davie, & Samaranayaka, 2010), this may be because the overwhelming majority of hospitalized assault are men aged 15 to 35 years (Langley & Gulliver, 2012). There are also a number of pregnant women who are victims of assault during pregnancy, for whom the subsequent risk of a fatal assault is substantially increased (Kady, Gilbert, Xing, & Smith, 2005), who will not be identified if their principal diagnosis is recorded using a code taken from “Pregnancy” chapter of ICD-10 (e.g., “O02.0, Threatened abortion”).

In addition, ICD-10-AM coding rules require certainty when an “assault” external cause of injury code is recorded (National Centre for Classification in Health, 2002). As such, where there is doubt about the cause of injury (where there were no witnesses or the injured person chooses not to divulge the cause of the injury), assault will not be recorded. The frequency of assault-related injury discharges reported will be an underestimate. While this is a limitation of the ICD coding scheme, ICD does allow for the recording of assaultive events at different locations and during various activities. For example, if an assault on a sports field is reported as such during a hospital event, it will be captured. These limitations and strengths of the ICD coding scheme are also inherent in the method proposed by Estrada.

We have not investigated the validity of our indicators for different age groups or ethnic groups in the New Zealand population. Up to three ethnicities may be recorded in the NMDS. For the purposes of reporting, a prioritized reporting scheme is used, whereby those who have reported Maori (the indigenous people of New Zealand) in at least one ethnicity field are recorded as Maori; followed by Pacific Island ethnicities; Asian; Middle Eastern, Latin American, and African. Those who are only recorded as European are reported as such. With the exception of Maori and European ethnicities, the remaining are not reported in sufficiently high frequencies to allow investigation of the validity of these indicators. The serious nonfatal assault indicators are reported for Maori (Statistics New Zealand, 2011a). However, although out of the scope of this investigation, there are issues associated with the use of ethnicity as recorded in the NMDS (Harris et al., 2007).

The release of statistics relating to crime is of great public interest, in New Zealand as it is in other countries (Weatherburn, 2011). In April 2012, the New Zealand police released crime figures indicating that there had been a 4.8% reduction in reported crime in 2011 compared with 2010 (New Zealand Police, 2012a). The reason for such reductions should be considered carefully, however, before jumping to conclusions. Part of this recorded reduction was a 22% decrease in crime in the Canterbury region, which had been affected by a significant earthquake in February 2011. The recorded reduction was made up of “a sudden large decrease in Theft and Property Damage, with offences at the less serious end of the spectrum reducing the most. Although small by value, these offences are typically large by volume” (New Zealand Police, 2012b). In addition, there was a 14.9% increase in reported sexual assault and related offenses, an increase largely attributed to “an increase in the number of offences being reported rather than an increase in the number of offences” (New Zealand Police, 2012b). That they can be so dramatically influenced by factors which may be unrelated to the incidence of crime in the community underscores how inappropriate crime statistics are for monitoring the incidence of assault over time.

Estrada correctly points out that there are major limitations with the use of crime statistics to monitor trends in assault. He suggested that although hospital data suffers similar limitations there are solutions to these limitations. As we have demonstrated here, his solutions will not result in the derivation of valid indicators in trends in assault. Our proposed solution will. The solution we have proposed in this investigation has applicability to other countries that have a central repository for hospital discharge data. As highlighted earlier, ICD-10 is used in a large number of developed countries for recording the nature and circumstances of hospital event. Where ICD-10 is not being employed, an earlier version, ICD-9 is still in existence. This makes the use of the case definition for serious nonfatal hospitalized assault, defined in this investigation, applicable to all countries that employ the ICD coding scheme. Country specific DSPs are achievable using the methods we have employed. Alternatively, the International Collaborative Effort on Injury Statistics is in the process of developing international DSPs which may be able to be employed for countries using ICD-10 (Gedeborg et al., 2012).

To adequately describe the impact of assault on a community, it is important that valid and reliable indicators are available. Indicators for which a minimal number of caveats are required, that are minimally influenced by extraneous factors and that allow lay members of the public to understand trends at a glance. The methods outlined in this article describe the use of routinely available administrative data in existence in the majority of developed countries. Use of these methods would allow the production of indicators to measure trends in serious nonfatal assaultive injury that would not be unduly influenced by extraneous factors.

Footnotes

Acknowledgements

The authors gratefully acknowledge the constructive comments received from Gabrielle Davie on an earlier draft of this manuscript.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by a number of independent projects. These projects were funded by the Accident Compensation Corporation of New Zealand; Statistics New Zealand; and the University of Otago.

Notes

Author Biographies

Pauline Gulliver, was employed as a Research Fellow at the Injury Prevention Research between 2001 and 2012. She completed her PhD in Human Nutrition at Otago University in 1998 and has worked for the Cancer Society of New Zealand and Kings College London before returning to Otago University. She has worked on epidemiological investigations of childhood injuries in the home, the surveillance and description of sports injuries, predictors of traffic crash involvement, and the development and evaluation of indicators of injury incidence. She is currently employed at the New Zealand Family Violence Clearinghouse, University of Auckland.

John Langley, has been involved in injury prevention research since 1978, and received his PhD from the University of Otago in 1985. He has written numerous articles in academic, professional and lay publications on a wide range of injury prevention issues. He tends to use quantitative research methods, especially epidemiological. He is an active advocate for improvements in injury prevention policy and practice.

Colin Cryer, trained in statistics originally, and taught biostatistics at the Wellington School of Medicine during the 1980s. He has been involved in injury prevention research since 1981. His occupational injury work includes involvement in the two work-related fatal injury studies in New Zealand—carried out in the 1980s and the 1990s, as well as projects relating to the cause and prevention of injury among forestry workers and on farms. As well as the occupational injury projects, he has been involved in research aimed at improving the validity of indicators and surveillance methods. The most recent of these resulted in a report to the New Zealand Injury Prevention Strategy (NZIPS) Secretariat on developing valid injury outcome indicators.

References

Accident Compensation Corporation. (2003). New Zealand Injury Prevention Strategy. Wellington, New Zealand: Author.

Cryer

Gulliver

Davie

de Graaf

Langley

Boufous

Steiner

C. A.

(2011). Empirical validation of the New Zealand Injury Prevention Strategy indicators: The identification of ICD diagnoses associated with a high probability of inpatient hospital admission. Wellington, New Zealand: Injury Prevention Research Unit (University of Otago).

Cryer

Gulliver

Langley

Davie

(2010). Is length of stay in hospital a stable proxy for injury severity? Injury Prevention, 16, 254-260.

Cryer

Gulliver

Langley

Davie

Samaranayak

(2010). The New Zealand Injury Prevention Strategy serious injury outcome indicators: Does the case definition of serious non-fatal injury miss a material number of serious injury cases? In System

T. O. S.

(Ed.), Official statistics research series (p. 1-38). Wellington, New Zealand: University of Otago.

Cryer

Langley

(2006). Developing valid indicators of injury incidence for “all injury”. Injury Prevention, 12, 202-207.

Cryer

Langley

Stephenson

(2004). Developing valid injury outcome indicators: A report for the New Zealand Injury Prevention Strategy. Dunedin, New Zealand: University of Otago.

Davie

Langley

Samaranayaka

Wetherspoon

M. E.

(2008). Accuracy of injury coding under ICD-10-AM for New Zealand public hospital discharges. Injury Prevention, 14, 319-323.

Estrada

(2001). Juvenile violence as a social problem: Trends, media attention and societal response. British Journal of Criminology, 41, 639-655.

Estrada

(2006). Trends in violence in Scandanavia according to different indicators. British Journal of Criminology, 46, 486-504.

10.

Federal Bureau of Investigation (Producer). (2010). Crime in the United States, 2009. Retrieved from http://www2.fbi.gov/ucr/cius2009/offenses/violent_crime/index.html

11.

Gedeborg

Warner

Chen

L.-H.

Gulliver

Robitaille

Bauer

Langley

(2012). Internationally comparable diagnosis-specific survival probabilities for calculation of the ICD-10 based injury severity score (ICE-ICISS). Journal of Trauma, Manuscript submitted.

12.

Gulliver

Cryer

Davie

(2009). Investigation of the provisional status of the NZIPS serious non-fatal assault and self-harm indicators. Dunedin, New Zealand: Injury Prevention Research Unit, University of Otago.

13.

Gulliver

Cryer

Davie

(2010). A chartbook of the New Zealand Injury Prevention Strategy serious injury outcome indicators: 1994-2009. Wellington, New Zealand: Injury Prevention Research Unit, Dunedin School of Medicine.

14.

Harris

Purdie

Robson

Wright

Zhang

Baker

(2007). Appendix 3: Estimating Maori hospitalisations and cancer registrations. In Robson

B. H.

Harris

(Eds.), Hauora: Maori standards of health IV. A study of the years 2000-2005 (pp. 249-259). Wellington, New Zealand: Te Ropu Rangahau Hauora a Eru Pomare.

15.

Kady

D. E.

Gilbert

W. M.

Xing

Smith

L. H.

(2005). Maternal and neonatal outcomes of assaults during pregnancy. Obstetric Gynecology, 105, 357-363.

16.

Langley

(2006). Emergency department data suffer similar threats to validity as police data. Injury Prevention, 12, 208.

17.

Langley

Gulliver

(2012). A decade of serious non-fatal assault in New Zealand. New Zealand Medical Journal, 125, 65-76.

18.

Langley

Stephenson

Cryer

(2003). Measuring road traffic safety performance: monitoring trends in non-fatal injury. Traffic Injury Prevention, 4, 291-296.

19.

Langley

Stephenson

Cryer

Borman

(2002). Traps for the unwary in estimating person based injury incidence using hospital discharge data. Injury Prevention, 8, 332-337.

20.

Ministry of Health. (2012). National minimum dataset (hospital events). Retrieved from New Zealand Government http://www.health.govt.nz/nz-health-statistics/national-collections-and-surveys/collections/national-minimum-dataset-hospital-events

21.

National Centre for Classification in Health. (2000a). ICD-10-AM tabular list of diseases, volume 1 of The International Statistical Classification of Diseases and related health problems, 10th revision, Australian modification. Sydney, Australia: Author.

22.

National Centre for Classification in Health (Ed.). (2000b). International Statistical Classification of Diseases and health related problems, 10th revision, Australian modification (2nd ed., Vol. 1). Sydney, Australia: Author.

23.

National Centre for Classification in Health (Ed.). (2002). International Statistical Classification of Diseases and health related problems, 10th revision, Australian modification (3rd ed., Vol. 1). Sydney, Australia: Author.

24.

New Zealand Police. (2012a). Deputy Commissioner Mike Bush on record fall in recorded crime. Press Release.

25.

New Zealand Police (Producer). (2012b, April 18). Questions and answers: 2011 Calendar year crime statistics’. Retrieved from http://www.police.govt.nz/question-and-answers-2011-calendar-year-crime-statistics

26.

Oxford Dictionaries (Producer). (2012, November 19). Violence. Retrieved from http://oxforddictionaries.com/definition/english/violence

27.

Shepherd

Sivarajasingam

(2005). Injury research explains conflicting violence trends. Injury Prevention, 11, 324-325.

28.

Statistics New Zealand. (2010). Protocol for government agency reporting on injury incidence in New Zealand. Wellington, New Zealand: Statistics New Zealand. Retrieved from http://www.stats.govt.nz/browse_for_stats/health/injuries/injury-reporting-protocol.aspx

29.

Statistics New Zealand. (2011a). Serious injury outcome indicators for Maori: 1996-2010. Wellington, New Zealand: Author.

30.

Statistics New Zealand. (2011b). Serious injury outcome indicators: 1994-2010. Wellington, New Zealand: Author.

31.

Stephenson

Langley

Civil

(2002). Comparing measures of injury severity for use with large databases. Journal of Trauma, 53, 326-332.

32.

Violence Prevention Alliance (Producer). (2012, March 5). Definition and typology of violence. Retrieved from http://www.who.int/violenceprevention/approach/definition/en/index.html

33.

Weatherburn

(2011). Uses and abuses of crime statistics. Crime and Justice Bulletin (pp. 1-16). Sydney, Australia: NSW Bureau of Crime Statistics and Research.